gms | German Medical Science

49. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie (gmds)
19. Jahrestagung der Schweizerischen Gesellschaft für Medizinische Informatik (SGMI)
Jahrestagung 2004 des Arbeitskreises Medizinische Informatik (ÖAKMI)

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie
Schweizerische Gesellschaft für Medizinische Informatik (SGMI)

26. bis 30.09.2004, Innsbruck/Tirol

Creative and Innovative Statistics in Clinical Research and Development

Meeting Abstract (gmds2004)

Suche in Medline nach

Kooperative Versorgung - Vernetzte Forschung - Ubiquitäre Information. 49. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie (gmds), 19. Jahrestagung der Schweizerischen Gesellschaft für Medizinische Informatik (SGMI) und Jahrestagung 2004 des Arbeitskreises Medizinische Informatik (ÖAKMI) der Österreichischen Computer Gesellschaft (OCG) und der Österreichischen Gesellschaft für Biomedizinische Technik (ÖGBMT). Innsbruck, 26.-30.09.2004. Düsseldorf, Köln: German Medical Science; 2004. Doc04gmds003

Die elektronische Version dieses Artikels ist vollständig und ist verfügbar unter: http://www.egms.de/de/meetings/gmds2004/04gmds003.shtml

Veröffentlicht: 14. September 2004

© 2004 Maurer.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen (http://creativecommons.org/licenses/by-nc-nd/3.0/deed.de). Er darf vervielf&aauml;ltigt, verbreitet und &oauml;ffentlich zug&aauml;nglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.


Text

Introduction

Innovation and creativity are not usually attributes that first come to one's mind when characterizing the qualities of a statistician. In the early days of statistics in the pharmaceutical industry and medical university departments, statisticians were mostly seen as professionals in a service function. The statistician was somebody who helped to analyze and document data from trials and experiments and who was, at best, only marginally involved in the planning stage [1], [2].

In this presentation it will be shown that the expectations and the scope of the tasks have changed greatly. One important change has been the realization (at least in some institutions) that statisticians can and should play a primary role in the design and analysis of studies and Research and Development (R&D) programs. Statisticians are also increasingly recognized as important contributors to - and sometimes the drivers of - innovation. For the managers, whether in a scientific institution or in industry, the bottom line is to deliver promising new treatment interventions by (i) providing convincing evidence of effectiveness, (ii) delivering sufficient confidence in safety, while minimizing the time to reimbursement in a cost-effective manner.

Statistics, Regulations and Innovation

Biostatistics in the Pharmaceutical Industry is highly regulated, particularly in areas primarily related to the submission of new drug applications (NDAs) for approval by health authorities (HA). The introduction of novel statistical methodology for NDAs is generally received by the HAs with caution. New methodologies or approaches are then often not proposed due to a perceived concern that the health authorities may not accept them. Indeed frequently due to lack of resources and time pressure, "previously accepted approaches" are used.

In March 2004 the US Food and Drug Administration (FDA) issued a "white paper" entitled "Stagnation / Innovation: Challenge and Opportunity on the Critical Path to New Medical Products". It states that today's revolution in biomedical science has raised new hope for the prevention, treatment, and cure of serious illnesses. However over the last few years, the number of new drug and biologic applications submitted to the FDA has declined significantly.

Within this white paper the potential causes for this recent decline, instead of the expected acceleration, in innovative medical therapies is assessed. One cause described is that the current medical product development path is becoming increasingly challenging, inefficient, and costly. As a consequence innovation will continue to stagnate or decline and the biomedical revolution may not deliver its promise of better health.

This white paper does not directly mention the role statisticians could play in making the drug development cycle more efficient and effective. However, there are in fact many possibilities where statistics and biometry in general can contribute in a creative and innovative way to achieve efficient and effective drug development. For illustration, examples are primarily used from areas where the statistical methodology group at Novartis Pharma AG together with consultants from academia were involved and where novel approaches have been or are planned to be implemented. Clearly there are many similar examples from other areas or institutions that could be mentioned.

Areas and examples of novel statistical approaches

Proof of concept (PoC) trials : PoC trials are the 'gatekeepers' between Research and (full) Development. Learning, quantification and verification of "proof" is the primary objective for these trials. Designing and analyzing them in the same way as phase III trials is generally inappropriate. Bayesian and adaptive designs are particularly well suited to PoC as they facilitate learning during data accrual and decision making. The statistical methodology to be applied needs to be tailored to each particular case and requires a firm knowledge of Bayesian techniques, creativity and interdisciplinary understanding, interest and cooperation.

Early safety signal detection : The cause for costly failures in phase III are often safety issues that have not been recognized in earlier trials. Detecting safety problems late in development puts an unnecessary number of patients at risk and is a major loss in (future) financial resource. The development of new statistical approaches, such as mixture models combined with data mining to define good predictors for future clinical manifestations of adverse events is therefore one route that could help to reduce the problem.

Adaptive Seamless Designs : Adaptive and seamless clinical trials allow learning from information accrued during the trial in a confirmatory framework, e.g., combining phase IIb and III trials in one single trial. They have the potential to shorten the development time and save costs on the critical path of the drug development cycle. In the last 5 years, the statistical methodology has been developed that allows to plan and analyze such trials [3], [4] in a way that should make them acceptable to regulatory authorities with respect to the attributes usually requested of pivotal phase III trials: control of probability of false positive error rate, integrity and selection bias adjustment .

Randomization schemes and optimizing drug supply : In a trial with many treatment arms (e.g. a factorial design with multiple dose levels for each factor) central randomization and drug supply can lead to a waste of unused drugs. Randomization within a center avoids this but can lead to considerable imbalance of treatment assignment and hence loss of power. Statistics can provide the tools to quantify the problem and propose randomization schemes that improve overall performance.

The 2 trial paradigm : A standard convention in seeking approval from HAs for new treatments is the need to demonstrate consistency (by individual rejection of the null-hypothesis) of results in (at least) two well-controlled trials - the "two-trials rule". Frequently, the two trials are very similar and more efficient use of these data can be found while providing adequate proof of consistency and overall type one error control [5], [6]

Dose finding :Dose finding is intrinsically a difficult task in clinical R&D [5], with evidence suggesting that many drugs are initially marketed at too high a dosage. Frequently, dose finding trials have been analyzed using a multiple comparison procedure. Modeling, on the other hand, provides a flexible tool to deliver a richer understanding of the underlying dose-response shape, but requires one to make an assumption for the functional form of the dose-response relationship. Statistical methods can be formulated such that these two concepts (control of false positive rate and estimation) are well characterized or fully integrated.

Handling of missing values single vs. multiple imputation : Single imputation (e.g., last observation carried forward, worst/best case) have historically been implemented extensively in clinical trials data analysis. The resulting "complete" data set is then analyzed as if the information had been observed. With the advent of easily amenable software releases multiple imputation strategies can be formulated, implemented and appropriately analyzed while avoiding the rather ad-hoc rules underpinning many single imputation schemes.

Other areas where innovative statistics improves the efficiency that will be covered briefly are new data mining tools and the thoughtful use (or avoidance) of dichotomization or other information reducing approaches.

Conclusion

Statisticians have always contributed to progress in Clinical R&D inventing and devising novel solutions and methods for analyzing and interpreting data; the abundance of papers published in professional journals are proof of this. We must however, not only solve problems as they arise, but look for opportunities and take the initiative (and some risk), if we want to have an interesting and fulfilling job.


Acknowledgements

I would like to thank my colleagues M. Branson, P. Gallo, J. Maca, J. Pinheiro, G. Rosenkranz and J.L. Steimer for their support and contributions to the examples.


References

1.
Grieve AP. Do Statisticians Count? A Personal View, Pharmaceutical Statistics. 2002; 1: 35 - 43.
2.
Morgan D. et al Qualified Statisticians in the European Pharmaceutical Industry: Report of a European Federation of Statisticians in the Pharmaceutical Industry (EFSPI) Working Group DIAJ 1999; 33: 407-415
3.
Bauer P, Köhne K. Evaluation of experiments with adaptive interim analysis. Biometrics 1994; 50: 1029 - 1041
4.
Hommel G. Adaptive Modifications of Hypotheses After an Interim Analysis, Biometrical Journal 2001; 5:
5.
Senn S.(1997) Statistical Issues in Drug Development Wiley: Chichester
6.
Maca J, Gallo P, Branson M, Maurer W. Reconsidering some aspects of the two-trial paradigm, Journal of Biopharmaceutical Statistics 2002; 12: 107 - 109